Machine learning applications to sports event prediction and detection of match-fixing in sports betting markets

机器学习被成功地用于许多分类和预测目的，但最先进的体育预测不愿适当地采用现代机器学习方法，如深度神经网络。相反，统计模型通常被用于体育预测，而且经常被发现是更好的。例如，英格拉姆为网球提出的基于点的贝叶斯分层模型[1]，以及Dixon和Coles[2]的开创性论文提出了足球的泊松回归模型。尽管如此，最近的一些进展表明，机器学习如何在体育预测中取得成功。Wilkens[3]在2021年的一篇综述中总结了网球中的机器学习预测方法，并主张进行进一步的研究，因为这些策略中的许多仍然存在缺陷和不准确，在投注应用中的无利可图就证明了这一点。

Machine learning is successfully used for many classification and prediction purposes, but state-of-the-art sports prediction is reluctant to properly embrace modern machine learning methods such as deep neural networks. Instead, statistical models are generally favoured for sports prediction, and have often been found to be superior. For example, the point-based Bayesian hierarchical model proposed by Ingram [1] for tennis, and the seminal paper by Dixon and Coles [2] that proposed a Poisson regression model for football. Despite this, there are recent advancements that demonstrate how machine learning can be successful in sports prediction. A 2021 review by Wilkens [3] summarizes the machine learning prediction methods in the case of tennis and argues for further research since many of these strategies are still flawed and inaccurate, illustrated by their unprofitability in betting applications.